• http://www.onlinematters.com Arthur Coleman

    Excellent article. I love how you took the analysis to a first integral in order to show how the bimodal distribution turns into a smooth logarithmic function when cumulative sales are considered. As a retailer, you could also use this same analysis to look at the impact of merchandising (e.g. BOGO, 20% off one day sale) and then can more accurately predict sales from merchandising of various types.

    Now the hard question: what if my first visit is from a click on Google and my second from a click on Bing? How would you go about dealing with the data to get the same accurate analysis? Now you have two cookies and two apparently different people. We have figured this out through our internal analytics, but not sure we have still got it right. Tough problem to solve.

  • zwelling

    I would add a third point to your list of caveats: you are assuming your retailer has perfect information. We’ve learned, in practice, that even with the longer cookie window you advise, information gets lost to cookie deletion and cross-browser research/shopping. Our customers see more conversions later in the attribution window than your data would suggest. One quick analysis we ran indicated that 14% of revenue delivered to a client of ours after the first day would’ve been misallocated to incorrect (more recent) events were it not for our proprietary tracking system.

    Jeff Zwelling CEO Convertro.com